TABLE OF CONTENTS (contd)

Adversaries corrupt RAG stores and long-term memory to persistently warp future agent reasoning

Weak agent authentication allows message spoofing, replay attacks, and goal manipulation across networks

A single fault propagates through autonomous agents compounding into system-wide outages

Agents exploit automation bias to manipulate humans into approving harmful or fraudulent actions

Malicious or misaligned agents deviate from their intended function, sabotaging workflows from within